# US 2025/0168197 A1 · Ai-supported network telemetry using data processing unit

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0168197 A1
- Document ID: US20250168197A1
- Status: Published application
- Kind: A1
- Application number: 19/029,733
- Filed: 2025-01-17
- Published: 2025-05-22
- Earliest claimed: 2021-12-14
- Assignees: Mellanox Technologies, Ltd.
- Applicants: Mellanox Technologies, Ltd.
- Inventors: Dotan Finkelshtien; Alexander Bilkovskii; Roni Bar Yanai; Juan Jose Vegas Olmos
- Web page: https://patentgrep.com/patent/US20250168197A1

## CPC

- H04L 63/1408: {by monitoring network traffic (monitoring network traffic per se H04L43/00)}
- H04L 63/1466: {Active attacks involving interception, injection, modification, spoofing of data unit addresses, e.g. hijacking, packet injection or TCP sequence number attacks}
- G06N 20/00: Machine learning
- H04L 63/0245: {Filtering by information in the payload}
- H04L 63/1416: {Event detection, e.g. attack signature detection}
- H04L 63/1425: {Traffic logging, e.g. anomaly detection}
- H04L 63/1441: {Countermeasures against malicious traffic (countermeasures against attacks on cryptographic mechanisms H04L9/002)}
- H04L 63/166: {at the transport layer}
- H04L 63/20: {for managing network security; network security policies in general (filtering policies H04L63/0227)}
- H04L 69/22: Parsing or analysis of headers
- H04L 63/145: {the attack involving the propagation of malware through the network, e.g. viruses, trojans or worms}
- H04L 63/1458: {Denial of Service}
- H04L 63/1483: {service impersonation, e.g. phishing, pharming or web spoofing (detection of rogue wireless access points H04W12/12)}
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A device receives a packet from a local network. The packet may be directed toward a cloud computing resource. The device determines that the packet is associated with a new packet flow. In response to determining that the packet is associated with the new packet flow, the device provides one or more packets from the new packet flow to a machine learning model for packet inspection. The device receives an output from the machine learning model and routes the new packet flow based on the output received from the machine learning model. The output indicates whether or not the new packet flow is associated with a network attack.

## Claims

### Claim at position 1

1-20. (canceled)

### Claim 21

A system, comprising:
a network interface to receive a packet flow destined for a cloud computing resource;
one or more circuits to:
determine whether a source of the packet flow has previously communicated with the cloud computing resource;
provide, when the source of the packet flow has not previously communicated with the cloud computing resource, a first packet of the packet flow to a protocol stack comprising two or more layers of filtering logic that detect whether the packet flow is associated with a network attack or not associated with a network attack; and
send, when the source of the packet flow has previously communicated with the cloud computing resource, the packet flow to the cloud computing resource while bypassing the two or more layers of filtering logic.

### Claim 22 (depends on 21)

The system of claim 21, wherein the filtering logic comprises pre-configured attack-detection rule sets.

### Claim 23 (depends on 22)

The system of claim 22, wherein the pre-configured attack-detection rule sets comprise one or more of a signature, a data pattern, or a network attack signature.

### Claim 24 (depends on 22)

The system of claim 22, wherein, when the first packet does not match one of the pre-configured attack-detection rule sets, the one or more circuits are to forward the first packet of the packet flow to a machine learning model for packet inspection.

### Claim 25 (depends on 22)

The system of claim 22, wherein, when the first packet does not match one of the pre-configured attack-detection rule sets, the one or more circuits are to forward at least two packets, including the first packet, of the packet flow to a machine learning model for packet inspection.

### Claim 26 (depends on 21)

The system of claim 21, wherein, when the filtering logic detects that the packet flow is not associated with a network attack, the one or more circuits are to send other packets in the packet flow to the cloud computing resource via an offload path that bypasses the two or more layers of filtering logic.

### Claim 27 (depends on 26)

The system of claim 26, wherein the offload path caries the other packets at full wire speed.

### Claim 28 (depends on 21)

The system of claim 21, wherein the first packet of the packet flow is at a beginning of the packet flow.

### Claim 29 (depends on 21)

The system of claim 21, further comprising:
the cloud computing resource.

### Claim 30 (depends on 21)

The system of claim 21, wherein the two or more layers of filtering logic are configured to detect the network attack as one of a Distributed Denial of Service (DDOS) attack, a cryptominer attack, a broken access control, a security misconfiguration, an injection, a phishing attack, a malware attack, a ransomware attack, a cross-site scripting (XSS) attack, a sensitive data exposure, an information leakage, a cryptojacking, a fraudulent e-mail transmission, a botnet, a malicious insider attack, or a social profile engineering attack.

### Claim 31

A device, comprising:
a protocol stack comprising two or more layers of filtering logic that detect network attacks; and
one or more circuits to:
receive a packet flow destined for a cloud computing resource;
identify a source of the packet flow based on a header of a first packet in the packet flow;
determine whether the source has previously communicated with the cloud computing resource;
provide, when the source of the packet flow has not previously communicated with the cloud computing resource, the first packet of the packet flow to the two or more layers of filtering logic to detect whether the packet flow is associated with a network attack or not associated with a network attack; and
send, when the source of the packet flow has previously communicated with the cloud computing resource, the packet flow to the cloud computing resource while bypassing the two or more layers of filtering logic.

### Claim 32 (depends on 31)

The device of claim 31, wherein the filtering logic comprises pre-configured attack-detection rule sets.

### Claim 33 (depends on 32)

The device of claim 32, wherein the pre-configured attack-detection rule sets comprise one or more of a signature, a data pattern, or a network attack signature.

### Claim 34 (depends on 32)

The device of claim 32, wherein, when the first packet does not match one of the pre-configured attack-detection rule sets, the one or more circuits are to forward the first packet of the packet flow to a machine learning model for packet inspection.

### Claim 35 (depends on 32)

The device of claim 32, wherein, when the first packet does not match one of the pre-configured attack-detection rule sets, the one or more circuits are to forward at least two packets, including the first packet, of the packet flow to a machine learning model for packet inspection.

### Claim 36 (depends on 31)

The device of claim 31, wherein, when the filtering logic detects that the packet flow is not associated with a network attack, the one or more circuits are to send other packets in the packet flow to the cloud computing resource via an offload path that bypasses the two or more layers of filtering logic.

### Claim 37 (depends on 36)

The device of claim 36, wherein the offload path caries the other packets at full wire speed.

### Claim 38 (depends on 31)

The device of claim 31, wherein the first packet of the packet flow is at a beginning of the packet flow.

### Claim 39 (depends on 31)

The device of claim 31, wherein the two or more layers of filtering logic are implemented by respective layers of the protocol stack.

### Claim 40

A system, comprising:
a machine learning model to detect network attacks; and
one or more circuits to:
receive a packet flow destined for a cloud computing resource;
determine whether a source of the packet flow has previously communicated with the cloud computing resource;
provide, when the source of the packet flow has not previously communicated with the cloud computing resource, a first packet of the packet flow to at least one of the machine learning model and a protocol stack comprising two or more layers of filtering logic that detect whether the packet flow is associated with a network attack or not associated with a network attack; and
send, when the source of the packet flow has previously communicated with the cloud computing resource, the packet flow to the cloud computing resource while bypassing the two or more layers of filtering logic and the machine learning model.
